Fix pip module tests (#423)

Status: Ready

This fixes test failures related to module loading using `pip install
"mhkit[<module>]"` syntax.

There are some modules that have interdependencies with other modules,
i.e., loads module needs wave module code that are causing test
failures.

The root cause of test failures is specific packages that are not
specified in pyproject.toml for a specific module, but are called when
the tests are run.

The fix is to add lazy loading for the `wave.io` module and reconfigure
the modules in pyproject.toml to capture the a more correct dependency
list derived from running the tests.

Also the command that runs these tests was changed from `pip install -e
"mhkit[<module>]" .` to `pip install -e ".[<module>]"` because the tests
fail with `-e "mhkit"` but succeed with `-e "."`. I suspect that this
changes the way mhkit is installed, but I don't know what the specific
differences are.

---------

Co-authored-by: jmcvey3 <53623232+jmcvey3@users.noreply.github.com>
Co-authored-by: akeeste <akeeste@sandia.gov>

Author: 3 people

.github/workflows/main.yml

@@ -458,6 +458,7 @@ jobs:
     needs: [set-os, prepare-nonhindcast-cache]
     runs-on: ubuntu-latest
     strategy:
+      fail-fast: false
       matrix:
         module:
           [wave, tidal, river, dolfyn, power, loads, mooring, acoustics, utils]
@@ -478,15 +479,17 @@ jobs:
           path: ~/.cache/mhkit
 
       - name: Install system dependencies
+        if: matrix.module != 'river' || matrix.module != 'power' || matrix.module != 'utils' || matrix.module != 'loads'
         run: sudo apt-get install -y libhdf5-dev libnetcdf-dev
 
       - name: Install MHKiT with optional dependency
         run: |
           python -m pip install --upgrade pip
-          pip install "mhkit[${{ matrix.module }}]"
+          pip install -e ".[${{ matrix.module }}]"
           pip install pytest
 
       - name: Reinstall h5py and netCDF4 with system libraries
+        if: matrix.module != 'river' || matrix.module != 'power' || matrix.module != 'utils' || matrix.module != 'loads'
         run: "pip install --force-reinstall --no-binary=:all: h5py netCDF4"
 
       - name: Run tests for ${{ matrix.module }}

mhkit/wave/__init__.py

@@ -1,5 +1,37 @@
+import importlib
+
 from mhkit.wave import resource
-from mhkit.wave import io
 from mhkit.wave import graphics
 from mhkit.wave import performance
 from mhkit.wave import contours
+
+
+def __getattr__(name):
+    """
+    Lazy load the wave.io submodule using PEP 562 module-level __getattr__.
+
+    This defers importing heavy wave.io dependencies (rex, netCDF4, etc,) until
+    they are actually accessed, improving import time for users who don't need
+    all wave submodules, and avoiding import errors for users who have specified
+    module level installs that need wave module functions, but not wave.io functions.
+    """
+    if name == "io":
+        # This uses importlib.import_module() here, not "from mhkit.wave import io"
+        # because when Python executes getattr(), it looks for 'io' as an attribute of
+        # mhkit.wave. At this point in the module loading code 'io' doesn't exist yet and
+        # Python calls __getattr__('io') again. This triggers the same "from" statement,
+        # which calls __getattr__('io') again yielding a RecursionError.
+        #
+        # To fix this uses importlib.import_module("mhkit.wave.io") which loads the module directly
+        # using the  absolute path without doing attribute lookup on the parent.
+        #
+        # The statement "from mhkit.wave import io" is equivalent to:
+        #   io = getattr(mhkit.wave, 'io')
+        #
+        io = importlib.import_module("mhkit.wave.io")
+
+        # Cache the module so subsequent accesses bypass __getattr__ entirely
+        globals()[name] = io
+        return io
+
+    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")

pyproject.toml

@@ -27,6 +27,7 @@ dependencies = [
     "xarray>=2024.6.0",
     "matplotlib>=3.9.1",
     "pecos>=0.3.0",
+    "requests",
 ]
 
 [project.optional-dependencies]
@@ -38,20 +39,17 @@ wave = [
     "pytz",
     "NREL-rex>=0.2.63",
     "beautifulsoup4",
-    "requests",
     "bottleneck",
     "lxml"
 ]
 
 tidal = [
     "netCDF4>=1.7.1.post1",
-    "requests",
     "bottleneck"
 ]
 
 river = [
     "netCDF4>=1.7.1.post1",
-    "requests",
     "bottleneck",
 ]
 
@@ -66,7 +64,9 @@ power = [
 ]
 
 loads = [
- "fatpack"
+    "fatpack",
+    "statsmodels>=0.14.2",
+    "scikit-learn>=1.5.1",
 ]
 
 mooring = [